Return NotImplemented from comparisons with non-Ksuid objects

[gaoflow]

Problem

Ksuid.__eq__ asserts isinstance(other, self.__class__), so comparing a Ksuid against any other type raises AssertionError instead of returning False:

>>> from ksuid import Ksuid, KsuidMs
>>> k = Ksuid()
>>> k == None
AssertionError
>>> None in [k]
AssertionError
>>> KsuidMs() == Ksuid()
AssertionError

This breaks ordinary operations — ksuid == None, ksuid in some_list (list membership uses ==), comparing a Ksuid with a KsuidMs, and dict/set use with mixed types. It is also fragile under python -O, where assert is stripped and the same comparison instead raises AttributeError ('NoneType' object has no attribute '_uid'). __lt__ has the same problem.

Fix

Per the Python data model, rich comparison methods must return NotImplemented (never raise) when the other operand is of an unsupported type. __eq__ and __lt__ now return NotImplemented when other is not a Ksuid, so equality falls back to False and ordering raises a proper TypeError. Comparison between Ksuid and KsuidMs (both 20-byte values) now compares by raw bytes, consistent with __hash__.

Tests

Added test_eq_with_non_ksuid_does_not_raise, test_eq_by_raw_bytes, and test_ordering_with_non_ksuid_raises_type_error. The full suite passes (2021), and ty/ruff check/ruff format --check are clean on the ksuid package.

I worked on this with AI assistance under my direction and reviewed the change myself.

[svix-jbrown]

good catch, but I'm not sure I entirely agree with the logic for cross-class comparisons; I kind of feel like KsuidMs should only be equal to another KsuidMs, not to a Ksuid that happens to have the same internal byte representation.

def __eq__(self, other: object) -> bool:
    if not isinstance(other, Ksuid) or self.TIMESTAMP_LENGTH_IN_BYTES != other.TIMESTAMP_LENGTH_IN_BYTES:
        return NotImplemented
    return self._uid == other._uid

Along with a test confirming that a carefully-constructed Ksuid and KsuidMs can't overlap?

[gaoflow]

Good point, and agreed. I've updated __eq__ to guard on TIMESTAMP_LENGTH_IN_BYTES exactly as you suggested, so a Ksuid and a KsuidMs never compare equal even when their 20 raw bytes coincide.

Added test_eq_across_ksuid_subclasses_is_false, which constructs both types from the same bytes(range(20)), asserts their raw bytes are identical, and asserts they are not equal in either direction (while same-class equality still holds). Full suite is green.

[svix-jbrown]

Please also run uv run ruff format ksuid and commit the results.

[svix-jbrown]

Sounds good to me! Thanks for your contribution!